Google: Gemini 3.5 Live Translate — speech-to-speech in 70+ languages in real time

Google unveiled Gemini 3.5 Live Translate — a speech-to-speech translation system (converting spoken words directly into spoken words in another language) with just a few seconds of latency. Unlike previous solutions that supported only English, the new version covers 70+ languages and more than 2,000 language combinations in meetings and calls.

What is speech-to-speech translation?

Speech-to-speech — unlike classic text translation — captures speech in real time, translates it, and immediately delivers the translated audio to the other party. Gemini 3.5 Live Translate preserves the intonation, rhythm, and pitch of the original speaker’s voice, maintaining natural communication rather than the robotic tone delivered by older methods.

SynthID protection and availability

All generated audio content carries a SynthID watermark — Google’s standard for marking synthetic speech that enables subsequent authenticity verification and prevents deepfake audio misuse. The system is available in developer public preview via the Gemini Live API and Google AI Studio, while a private preview for Google Meet Enterprise is underway. A global rollout on the Google Translate app (Android and iOS) is already in progress.

Scale of deployment

Google’s own Google Translate processes more than one billion words every month, giving a sense of the infrastructure scale underpinning the new system. The ride-hailing platform Grab, which uses the Gemini Live API, records more than 10 million voice calls per month — a potential user base that can immediately benefit from real-time multilingual translation.

Availability on development platforms means developers can already integrate translation into their own applications while awaiting the broader public rollout.